The pipelining talk gives some good examples of where most the the instruction level parallelism comes from. http://millcomputing.com/docs/pipelining/

Most of the rest of the CPU design is really just to enable this.